Reading Notes

Dharmesh Shah on HubSpot, Culture as Product, and the Science of Zigging

Source: Dharmesh Shah on HubSpot, Culture as Product, and the Science of Zigging

Notes — Dharmesh Shah on HubSpot, Culture as Product, and the Science of Zigging

Four questions [Adler frame]

Q1 — What is it about? Dharmesh Shah (co-founder and CTO of HubSpot) reflects on 18 years of building a $30B company by systematically doing the opposite of conventional wisdom: SMB focus when everyone said enterprise, broad multi-product when everyone said focus on one thing, radical transparency including post-IPO, and no direct reports ever. The through-line is first-principles systematic thinking applied to decisions, culture, public speaking, and product complexity alike.

Q2 — How is it argued? Through a series of named frameworks and personal anecdotes. Dharmesh is an engineer by disposition — he codes custom software to measure laughs per minute at his own talks, names everything, and runs NPS on employees quarterly. The argument is structural: articulate the actual problem (not the solution), create binary defaults to avoid politics, impose systems and constraints rather than relying on culture words, and apply the same iterative product thinking to culture that you’d apply to software.

Q3 — Is it true? The Culture as Product framing is highly persuasive because it operationalises something usually treated as intangible. The second law analogy is a useful mental model though not a precise physics claim. The zig-while-others-zag advice is correct as a meta-principle but dangerous without the caveat he actually gives: limit the dimensions on which you’re contrarian, and make sure you have genuine conviction, not just contrarianness.

Q4 — What of it? Two immediately actionable ideas: (1) adopt flash tags to calibrate signal strength in any communication; (2) use culture NPS quarterly and treat employee feedback as bug reports. The broader insight — that culture is not preserved but iterated — inverts the common founder instinct and is probably the most important correction this episode offers.

Glossary

Culture as Product — Dharmesh’s core postulate: every company builds two products — one for customers and one for its team. Culture is the latter. Follows that you must iterate it, run NPS on it, file bug reports against it, and never freeze it.

Flash tags — A hashtag taxonomy for disambiguating signal strength in written communication: #fyi (no response expected), #suggestion (I’d do it but no response needed), #recommendation (I’ve researched this; please respond if you disagree), #plea (deeply held; I beg you). Not a mandate at any level. Published at flashtag.org.

LPM (Laughs Per Minute) — Metric from stand-up comedy applied to business keynotes. Dharmesh wrote custom software to score his own talks. Target for business keynotes: 1.2–1.5 LPM. Techniques: funny words must be the last words of the sentence (give audience time to react); squeeze multiple punchlines from a single story setup.

SoloWare — Software built for exactly one person, with no UI testing, no user promises, and trivial shutdown costs. Dharmesh has been building it for 30 years. The LPM analyser is an example.

Fight for simplicity / Second Law of Thermodynamics — Companies move through three phases: (1) survive, (2) grow without stagnating, (3) fight complexity. Complexity is not catastrophic — it’s slow and reliable death. Binary defaults (all-or-nothing transparency, pure seat lottery) are simpler than any middle case. Simplicity requires explicit fighting because well-intentioned people will always introduce entropy.

Third-order cost of features — First-order cost: development time. Second-order: maintenance. Third-order: dimensional complexity added to the entire business. Every new product forces every decision, chart, and org allocation to be sliced by one more dimension. The third-order cost dwarfs the first two.

4P framework — Idea evaluation heuristic: Potential (max outcome if successful), Probability (of success), Passion/Proximity (do you care or are you well-situated?), Prowess (unfair advantage). Apply in this order — never filter by probability before assessing potential.

High conviction, low consensus bets — Peter Thiel’s framing applied at HubSpot: you must be right about something other people think you’re wrong about, and stay right for a very long time. HubSpot’s bets: SMB focus (18 years), multi-product from year one (inverted “focus on one thing”), radical transparency (extended to designated-insider policy at IPO).

Debate, Decide, Unite — HubSpot’s softer variant of Amazon’s “disagree and commit.” Debate openly, make a single-owner decision (not necessarily the executive), then align fully — even the losing side.

Federal vs state law (culture) — Some culture elements are federal (core, inviolable: transparency, SOLVE for the long term); others are state (configurable per org: meeting hours, role-specific constraints). Org leaders can localise state-level culture as long as they don’t override federal.

Reverse gravity (SMB → enterprise) — In software, market pull always runs upward: bigger customers pay more, stay longer, have higher NPS. Every successful software company becomes enterprise eventually. The corollary is that staying SMB takes explicit resistance and creates a defensible moat from companies that already went up.

Culture as Product

Dharmesh did not set out to “build a culture.” Brian Halligan assigned the task with no instructions. Dharmesh treated it as an engineering problem: write a Python function that predicts a new hire’s success. What would the variables be?

He sent a survey and received visceral pushback — employees associated “culture work” with inauthentic corporate poster-making. The insight: culture already existed; the job was to articulate what was there, not impose something new.

The Culture Code deck (16 slides initially, eventually 128 pages and public) answered one question: what are the attributes of humans who succeed at HubSpot? It included aspirational items with “liner notes” — explicit flags that said, “this is not yet true, but we want it to be.” Those aspirations became self-fulfilling: new hires read them as facts and behaved accordingly.

Operating principles:

  • Culture NPS survey every quarter; results published to all (including raw comments)
  • Bug reports categorised and addressed in next all-minds (all-hands) meeting — some fixes committed, others marked “works as designed”
  • Core values = federal law (transparency, SOLVE, etc.); configurable norms = state law
  • Never freeze culture code; the needs of the “customer” (employees) change as the company scales

The transparency example: at IPO, their investment bankers introduced “designated insiders” — typically 5–6 people with access to all financials. Dharmesh and Brian discovered via N+1 induction that there is no legal maximum. They designated every employee a designated insider on day one of trading.

Flash tags

The megaphone problem: founders have opinions on everything; people over-index any founder comment as directive. IMHO is too weak; mandates are too strong. Flash tags fill the middle.

TagMeaningResponse expected?
#fyiPassing this alongNo
#suggestionI’d do this if I were youNo
#recommendationI’ve researched this; I’d do itYes — explain if you disagree
#pleaDeeply held; I beg youYes — please seriously consider

None are mandates. The plea is a soft mandate that almost never fires. Published at flashtag.org. Also useful outside founder context: any manager with a megaphone effect benefits from this precision.

Fight for Simplicity

The second law of thermodynamics states entropy increases in closed systems. Dharmesh applies this at every level: code, organisation, product, decisions.

Stage model for companies:

  • Stage 1: Don’t die
  • Stage 2: Don’t stagnate (stagnation = slow death)
  • Stage 3: Fight complexity (entropy kills eventually, more reliably than failure)

Mechanisms for imposing simplicity:

  • Binary defaults (everything transparent vs. nothing) are simpler than any threshold rule
  • Feature constraint: every new feature required removing one (early HubSpot product)
  • Seat lottery: random assignment every quarter, with local optimisations over time — eliminated every sq ft of office politics
  • System-imposed constraints outlast culture words at scale

SMB focus and simplicity: Freemium forces simplicity — the product has to deliver value without a sales team explaining it. A structural constraint, not a cultural aspiration.

The 4P Framework

Order matters:

  1. Potential — What could this be if it works? Scale of 0–10. (Don’t skip this.)
  2. Probability — Expected value = potential × probability. Only meaningful after you have the potential number.
  3. Passion / Proximity — Do you care? Are you close to the problem? Passion can be acquired. Proximity is often underrated.
  4. Prowess — Why you? Unfair advantage: existing code, market access, team. This modifies probability upward.

Common mistake: applying probability as a filter before estimating potential. Result: you throw out big ideas because they’re risky, without ever computing that 10% × $1B > 50% × $10M.

High Conviction, Low Consensus Bets

HubSpot’s three major zigs:

  1. SMB market: no successful software company had been built purely for SMB before HubSpot. Investors pushed for enterprise path throughout the IPO roadshow.
  2. Multi-product from year one: inverts every “focus on one thing” startup mantra. Justified by: HubSpot’s value proposition is integration, not category leadership in any individual tool. Heuristic: if you’re in the top three in a single category, you over-invested in it.
  3. Radical transparency: no titles (years 1–4ish), flat pay structure early, all-insider policy at IPO.

The rule: limit the dimensions on which you’re contrarian. You need at least one, but not too many. Know explicitly what the contrarian zig would be, even if you don’t take it.

Lean Into Strengths

Dharmesh has had zero direct reports at 7,000 employees. This was a founding-day decision: “I cannot become passively okay at management. I refuse to spend years of my life becoming passively okay at something.” The deal with co-founder Brian: no interim-manage if a VP quit. The result: Dharmesh has more fun at 7,000 people than he had at 70.

Talent vs skill: talent controls the slope of the learning curve and the ceiling, but almost everything is an acquirable skill. LPM for public speaking is the worked example: functional decomposition → pick one sub-skill per year → measure.

Connections to Wiki